cabac.h 13.9 KB
Newer Older
1 2 3
/*
 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
 *
4
 * This file is part of FFmpeg.
5
 *
6
 * FFmpeg is free software; you can redistribute it and/or
7 8 9 10
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
11
 * FFmpeg is distributed in the hope that it will be useful,
12 13 14 15 16
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
17
 * License along with FFmpeg; if not, write to the Free Software
18 19 20 21 22 23 24 25
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#ifndef AVCODEC_X86_CABAC_H
#define AVCODEC_X86_CABAC_H

#include "libavcodec/cabac.h"
#include "libavutil/attributes.h"
26
#include "libavutil/macros.h"
27
#include "libavutil/x86/asm.h"
28 29
#include "config.h"

30
#if   (defined(__i386) && defined(__clang__) && (__clang_major__<2 || (__clang_major__==2 && __clang_minor__<10)))\
31 32
   || (                  !defined(__clang__) && defined(__llvm__) && __GNUC__==4 && __GNUC_MINOR__==2 && __GNUC_PATCHLEVEL__<=1)\
   || (defined(__INTEL_COMPILER) && defined(_MSC_VER))
33 34 35 36 37
#       define BROKEN_COMPILER 1
#else
#       define BROKEN_COMPILER 0
#endif

38 39
#if HAVE_INLINE_ASM

40 41 42 43 44 45 46 47 48 49 50 51
#ifndef UNCHECKED_BITSTREAM_READER
#define UNCHECKED_BITSTREAM_READER !CONFIG_SAFE_BITSTREAM_READER
#endif

#if UNCHECKED_BITSTREAM_READER
#define END_CHECK(end) ""
#else
#define END_CHECK(end) \
        "cmp    "end"       , %%"REG_c"                                 \n\t"\
        "jge    1f                                                      \n\t"
#endif

52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95
#ifdef BROKEN_RELOCATIONS
#define TABLES_ARG , "r"(tables)

#if HAVE_FAST_CMOV
#define BRANCHLESS_GET_CABAC_UPDATE(ret, retq, low, range, tmp) \
        "cmp    "low"       , "tmp"                        \n\t"\
        "cmova  %%ecx       , "range"                      \n\t"\
        "sbb    %%rcx       , %%rcx                        \n\t"\
        "and    %%ecx       , "tmp"                        \n\t"\
        "xor    %%rcx       , "retq"                       \n\t"\
        "sub    "tmp"       , "low"                        \n\t"
#else /* HAVE_FAST_CMOV */
#define BRANCHLESS_GET_CABAC_UPDATE(ret, retq, low, range, tmp) \
/* P4 Prescott has crappy cmov,sbb,64bit shift so avoid them */ \
        "sub    "low"       , "tmp"                        \n\t"\
        "sar    $31         , "tmp"                        \n\t"\
        "sub    %%ecx       , "range"                      \n\t"\
        "and    "tmp"       , "range"                      \n\t"\
        "add    %%ecx       , "range"                      \n\t"\
        "shl    $17         , %%ecx                        \n\t"\
        "and    "tmp"       , %%ecx                        \n\t"\
        "sub    %%ecx       , "low"                        \n\t"\
        "xor    "tmp"       , "ret"                        \n\t"\
        "movslq "ret"       , "retq"                       \n\t"
#endif /* HAVE_FAST_CMOV */

#define BRANCHLESS_GET_CABAC(ret, retq, statep, low, lowword, range, rangeq, tmp, tmpbyte, byte, end, norm_off, lps_off, mlps_off, tables) \
        "movzbl "statep"    , "ret"                                     \n\t"\
        "mov    "range"     , "tmp"                                     \n\t"\
        "and    $0xC0       , "range"                                   \n\t"\
        "lea    ("ret", "range", 2), %%ecx                              \n\t"\
        "movzbl "lps_off"("tables", %%rcx), "range"                     \n\t"\
        "sub    "range"     , "tmp"                                     \n\t"\
        "mov    "tmp"       , %%ecx                                     \n\t"\
        "shl    $17         , "tmp"                                     \n\t"\
        BRANCHLESS_GET_CABAC_UPDATE(ret, retq, low, range, tmp)              \
        "movzbl "norm_off"("tables", "rangeq"), %%ecx                   \n\t"\
        "shl    %%cl        , "range"                                   \n\t"\
        "movzbl "mlps_off"+128("tables", "retq"), "tmp"                 \n\t"\
        "shl    %%cl        , "low"                                     \n\t"\
        "mov    "tmpbyte"   , "statep"                                  \n\t"\
        "test   "lowword"   , "lowword"                                 \n\t"\
        "jnz    2f                                                      \n\t"\
        "mov    "byte"      , %%"REG_c"                                 \n\t"\
96
        END_CHECK(end)\
97
        "add"OPSIZE" $2     , "byte"                                    \n\t"\
98
        "1:                                                             \n\t"\
99 100 101 102 103 104 105 106 107 108 109 110 111 112 113
        "movzwl (%%"REG_c") , "tmp"                                     \n\t"\
        "lea    -1("low")   , %%ecx                                     \n\t"\
        "xor    "low"       , %%ecx                                     \n\t"\
        "shr    $15         , %%ecx                                     \n\t"\
        "bswap  "tmp"                                                   \n\t"\
        "shr    $15         , "tmp"                                     \n\t"\
        "movzbl "norm_off"("tables", %%rcx), %%ecx                      \n\t"\
        "sub    $0xFFFF     , "tmp"                                     \n\t"\
        "neg    %%ecx                                                   \n\t"\
        "add    $7          , %%ecx                                     \n\t"\
        "shl    %%cl        , "tmp"                                     \n\t"\
        "add    "tmp"       , "low"                                     \n\t"\
        "2:                                                             \n\t"

#else /* BROKEN_RELOCATIONS */
114
#define TABLES_ARG NAMED_CONSTRAINTS_ARRAY_ADD(ff_h264_cabac_tables)
115 116
#define RIP_ARG

117
#if HAVE_FAST_CMOV
118
#define BRANCHLESS_GET_CABAC_UPDATE(ret, low, range, tmp)\
119 120 121 122 123 124
        "mov    "tmp"       , %%ecx     \n\t"\
        "shl    $17         , "tmp"     \n\t"\
        "cmp    "low"       , "tmp"     \n\t"\
        "cmova  %%ecx       , "range"   \n\t"\
        "sbb    %%ecx       , %%ecx     \n\t"\
        "and    %%ecx       , "tmp"     \n\t"\
125 126
        "xor    %%ecx       , "ret"     \n\t"\
        "sub    "tmp"       , "low"     \n\t"
127
#else /* HAVE_FAST_CMOV */
128
#define BRANCHLESS_GET_CABAC_UPDATE(ret, low, range, tmp)\
129 130 131 132 133 134 135 136 137 138 139 140 141
        "mov    "tmp"       , %%ecx     \n\t"\
        "shl    $17         , "tmp"     \n\t"\
        "sub    "low"       , "tmp"     \n\t"\
        "sar    $31         , "tmp"     \n\t" /*lps_mask*/\
        "sub    %%ecx       , "range"   \n\t" /*RangeLPS - range*/\
        "and    "tmp"       , "range"   \n\t" /*(RangeLPS - range)&lps_mask*/\
        "add    %%ecx       , "range"   \n\t" /*new range*/\
        "shl    $17         , %%ecx     \n\t"\
        "and    "tmp"       , %%ecx     \n\t"\
        "sub    %%ecx       , "low"     \n\t"\
        "xor    "tmp"       , "ret"     \n\t"
#endif /* HAVE_FAST_CMOV */

142
#define BRANCHLESS_GET_CABAC(ret, retq, statep, low, lowword, range, rangeq, tmp, tmpbyte, byte, end, norm_off, lps_off, mlps_off, tables) \
143 144 145
        "movzbl "statep"    , "ret"                                     \n\t"\
        "mov    "range"     , "tmp"                                     \n\t"\
        "and    $0xC0       , "range"                                   \n\t"\
146
        "movzbl "MANGLE(ff_h264_cabac_tables)"+"lps_off"("ret", "range", 2), "range" \n\t"\
147
        "sub    "range"     , "tmp"                                     \n\t"\
148
        BRANCHLESS_GET_CABAC_UPDATE(ret, low, range, tmp)                    \
149
        "movzbl "MANGLE(ff_h264_cabac_tables)"+"norm_off"("range"), %%ecx    \n\t"\
150
        "shl    %%cl        , "range"                                   \n\t"\
151
        "movzbl "MANGLE(ff_h264_cabac_tables)"+"mlps_off"+128("ret"), "tmp"  \n\t"\
152
        "shl    %%cl        , "low"                                     \n\t"\
153
        "mov    "tmpbyte"   , "statep"                                  \n\t"\
154
        "test   "lowword"   , "lowword"                                 \n\t"\
155
        " jnz   2f                                                      \n\t"\
156
        "mov    "byte"      , %%"REG_c"                                 \n\t"\
157
        END_CHECK(end)\
158
        "add"OPSIZE" $2     , "byte"                                    \n\t"\
159
        "1:                                                             \n\t"\
160 161 162 163
        "movzwl (%%"REG_c")     , "tmp"                                 \n\t"\
        "lea    -1("low")   , %%ecx                                     \n\t"\
        "xor    "low"       , %%ecx                                     \n\t"\
        "shr    $15         , %%ecx                                     \n\t"\
164 165
        "bswap  "tmp"                                                   \n\t"\
        "shr    $15         , "tmp"                                     \n\t"\
166
        "movzbl "MANGLE(ff_h264_cabac_tables)"+"norm_off"(%%ecx), %%ecx \n\t"\
167
        "sub    $0xFFFF     , "tmp"                                     \n\t"\
168 169 170 171
        "neg    %%ecx                                                   \n\t"\
        "add    $7          , %%ecx                                     \n\t"\
        "shl    %%cl        , "tmp"                                     \n\t"\
        "add    "tmp"       , "low"                                     \n\t"\
172
        "2:                                                             \n\t"
173

174
#endif /* BROKEN_RELOCATIONS */
175

176
#if HAVE_7REGS && !BROKEN_COMPILER
177 178 179 180
#define get_cabac_inline get_cabac_inline_x86
static av_always_inline int get_cabac_inline_x86(CABACContext *c,
                                                 uint8_t *const state)
{
181
    int bit, tmp;
182 183 184 185 186 187
#ifdef BROKEN_RELOCATIONS
    void *tables;

    __asm__ volatile(
        "lea    "MANGLE(ff_h264_cabac_tables)", %0      \n\t"
        : "=&r"(tables)
188
        : NAMED_CONSTRAINTS_ARRAY(ff_h264_cabac_tables)
189 190
    );
#endif
191 192

    __asm__ volatile(
193 194
        BRANCHLESS_GET_CABAC("%0", "%q0", "(%4)", "%1", "%w1",
                             "%2", "%q2", "%3", "%b3",
195 196 197 198 199
                             "%c6(%5)", "%c7(%5)",
                             AV_STRINGIFY(H264_NORM_SHIFT_OFFSET),
                             AV_STRINGIFY(H264_LPS_RANGE_OFFSET),
                             AV_STRINGIFY(H264_MLPS_STATE_OFFSET),
                             "%8")
200
        : "=&r"(bit), "=&r"(c->low), "=&r"(c->range), "=&q"(tmp)
201
        : "r"(state), "r"(c),
202
          "i"(offsetof(CABACContext, bytestream)),
203 204
          "i"(offsetof(CABACContext, bytestream_end))
          TABLES_ARG
205
          ,"1"(c->low), "2"(c->range)
206 207 208 209
        : "%"REG_c, "memory"
    );
    return bit & 1;
}
210
#endif /* HAVE_7REGS && !BROKEN_COMPILER */
211

212
#if !BROKEN_COMPILER
213 214 215 216 217
#define get_cabac_bypass_sign get_cabac_bypass_sign_x86
static av_always_inline int get_cabac_bypass_sign_x86(CABACContext *c, int val)
{
    x86_reg tmp;
    __asm__ volatile(
218 219
        "movl        %c6(%2), %k1       \n\t"
        "movl        %c3(%2), %%eax     \n\t"
220 221 222
        "shl             $17, %k1       \n\t"
        "add           %%eax, %%eax     \n\t"
        "sub             %k1, %%eax     \n\t"
223
        "cdq                            \n\t"
224 225 226 227 228 229
        "and           %%edx, %k1       \n\t"
        "add             %k1, %%eax     \n\t"
        "xor           %%edx, %%ecx     \n\t"
        "sub           %%edx, %%ecx     \n\t"
        "test           %%ax, %%ax      \n\t"
        "jnz              1f            \n\t"
230
        "mov         %c4(%2), %1        \n\t"
231 232 233 234
        "subl        $0xFFFF, %%eax     \n\t"
        "movzwl         (%1), %%edx     \n\t"
        "bswap         %%edx            \n\t"
        "shrl            $15, %%edx     \n\t"
235
#if UNCHECKED_BITSTREAM_READER
236
        "add              $2, %1        \n\t"
237
        "addl          %%edx, %%eax     \n\t"
238
        "mov              %1, %c4(%2)   \n\t"
239 240 241 242 243 244
#else
        "addl          %%edx, %%eax     \n\t"
        "cmp         %c5(%2), %1        \n\t"
        "jge              1f            \n\t"
        "add"OPSIZE"      $2, %c4(%2)   \n\t"
#endif
245
        "1:                             \n\t"
246
        "movl          %%eax, %c3(%2)   \n\t"
247

248 249 250 251
        : "+c"(val), "=&r"(tmp)
        : "r"(c),
          "i"(offsetof(CABACContext, low)),
          "i"(offsetof(CABACContext, bytestream)),
252
          "i"(offsetof(CABACContext, bytestream_end)),
253 254
          "i"(offsetof(CABACContext, range))
        : "%eax", "%edx", "memory"
255 256 257 258
    );
    return val;
}

259 260 261 262 263 264 265 266 267 268 269
#define get_cabac_bypass get_cabac_bypass_x86
static av_always_inline int get_cabac_bypass_x86(CABACContext *c)
{
    x86_reg tmp;
    int res;
    __asm__ volatile(
        "movl        %c6(%2), %k1       \n\t"
        "movl        %c3(%2), %%eax     \n\t"
        "shl             $17, %k1       \n\t"
        "add           %%eax, %%eax     \n\t"
        "sub             %k1, %%eax     \n\t"
270
        "cdq                            \n\t"
271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297
        "and           %%edx, %k1       \n\t"
        "add             %k1, %%eax     \n\t"
        "inc           %%edx            \n\t"
        "test           %%ax, %%ax      \n\t"
        "jnz              1f            \n\t"
        "mov         %c4(%2), %1        \n\t"
        "subl        $0xFFFF, %%eax     \n\t"
        "movzwl         (%1), %%ecx     \n\t"
        "bswap         %%ecx            \n\t"
        "shrl            $15, %%ecx     \n\t"
        "addl          %%ecx, %%eax     \n\t"
        "cmp         %c5(%2), %1        \n\t"
        "jge              1f            \n\t"
        "add"OPSIZE"      $2, %c4(%2)   \n\t"
        "1:                             \n\t"
        "movl          %%eax, %c3(%2)   \n\t"

        : "=&d"(res), "=&r"(tmp)
        : "r"(c),
          "i"(offsetof(CABACContext, low)),
          "i"(offsetof(CABACContext, bytestream)),
          "i"(offsetof(CABACContext, bytestream_end)),
          "i"(offsetof(CABACContext, range))
        : "%eax", "%ecx", "memory"
    );
    return res;
}
298
#endif /* !BROKEN_COMPILER */
299

300
#endif /* HAVE_INLINE_ASM */
301
#endif /* AVCODEC_X86_CABAC_H */